Polling in Iowa

The 2016 Iowa Caucuses are officially over, and the results are in: on the Democratic side, Hillary Clinton won, and for the Republicans, Ted Cruz. Comparing the results to political polls leading up to the caucus is important to reexamine their accuracy as well as how much weight polls should be given.

One particularly noteworthy poll came from the Des Moines Register/Bloomberg Politics and is conducted by Ann Selzer. Selzer’s polls became famous after accurately predicting Barack Obama’s 2008 win. This year, Selzer’s last poll before the caucus said Clinton led Sanders, 45 percent to 42 percent. The actual results were much closer: Clinton received 49.9 percent of the vote, Sanders received  49.6 percent.

While Selzer herself has said her polling methods are not infallible, it is interesting to note how her methods differ from other pollsters (many of which showed Sanders in the lead). The New York Times wrote up an examination of her methodology:

Most political polls get their data samples by dialing random telephone numbers and screening out unregistered voters. Selzer, on the other hand, calls phone numbers from voter registration files. Using voter files is advantageous because it essentially guarantees a demographically representative sample of registered voters. This method is riskier in Iowa, where people can register to vote on caucus night. The risk of random dialing polls, on the other hand, is that there are several potential biases that might skew results – more politically engaged people might be more likely to respond, people might misreport whether or not they’re registered, etc.

The next step is screening likely voters. Whereas mainstream media pollsters use a variety of questions like how enthusiastic the respondent is, whether or not the respondent has voted before, etc., Selzer's poll simply asks whether or not the respondent will vote. If the answer is "definitely" or "probably," the respondent is considered a likely voter. Using a variety of questions makes it more difficult to calculate turnout and easier to manipulate results; Selzer's single question is more straightforward. However, it is worth noting that voters tend to overestimate the likelihood of their attendance at the polls, and there are also voters that say they won't vote - but then do.
Erica Liao

1 comment :